SPATIAL INFERENCE AND CONSTRAINT SOLVING How to Depict Textual Spatial Descriptions from Internet

Today there are still many applications in the Internet, where the user is given a textual description of a spatial configuration (e.g. chat, e-mail or newsgroups). The user is asked to imagine the scene and to draw inferences. We present a new approach to generate depictions of such scenes. Besides of drawing spatial inferences, this leads to the problem of solving a system of complicated numerical constraints. In contrast to qualitative spatial reasoning, we use a metric description where relations between pairs of objects are represented by parameterized homogenous transformation matrices with numerical (nonlinear) constraints. We employ methods of machine learning in combination with a new algorithm for generating depictions from text including spatial inference.